Impact of Measurement Conditions on Classification of ADL using Surface EMG Signals

Studying Activities of Daily Living (ADL) is crucial for evaluating an individual's health and well-being, quantifying their functional status, and identifying limitations. The research in ADL has led to the development of various applications, such as prostheses, which can help individuals res...

Celý popis

Uložené v:
Podrobná bibliografia
Vydané v:2023 International Symposium on Image and Signal Processing and Analysis (ISPA) s. 1 - 6
Hlavní autori: Sagar Venna, Vidya, Turlapaty, Anish, Naidu, Surya
Médium: Konferenčný príspevok..
Jazyk:English
Vydavateľské údaje: IEEE 18.09.2023
Predmet:
Accuracy
ADL
Aggregates
Classification
Electromyography
feature extraction
measurement conditions
Performance evaluation
Signal processing
signals
Testing
Time-domain analysis
Training
Transfer learning
Wavelet domain
ISSN:1849-2266
On-line prístup:Získať plný text
Tagy: Pridať tag
Žiadne tagy, Buďte prvý, kto otaguje tento záznam!
Popis
Shrnutí:Studying Activities of Daily Living (ADL) is crucial for evaluating an individual's health and well-being, quantifying their functional status, and identifying limitations. The research in ADL has led to the development of various applications, such as prostheses, which can help individuals restore their ability to perform ADL and enhance their quality of life. Moreover, rehabilitation engineering studies ADL to create solutions and devices that improve the well-being of people with disabilities. The focus of this study is on EMAHA-DB4, a new collection of multi-channel surface electromyography (\mathbf{sEMG}) signals that is designed to evaluate ADL under a variety of measurement conditions. It consists of \mathbf{sEMG} signals from 10 subjects performing 8 unique and practical ADL that are essential. Each activity was performed in three different body postures and four arm positions to ensure a diverse range of measurement conditions. For constructing the framework, features derived from the time domain, frequency domain, wavelet domain, and Eigenvalues were utilized. The features are trained and tested with four different classical machine learning models and the performance is compared with the CNN Bi-LSTM, a hybrid deep learning architecture. It was found that the proposed ML framework can effectively classify under various conditions. In the aggregate scenario, the CNN Bi-LSTM achieved an average accuracy of 85.37% for classifying the activities versus arm positions and 82.1% versus body postures. Similarly, in the subject-wise scenario, CNN Bi-LSTM achieved an average accuracy of 84.4% for body conditions and 90.11% for arm conditions. In contrast, D NN consistently achieved higher accuracy across all conditions experiment in both the aggregate and subject-specific scenarios, with an average accuracy of 90.9% and 90.3% respectively. In these classification experiments, it is observed that the ML models are fairly robust to body postures but sensitive to arm positions.
ISSN:1849-2266
DOI:10.1109/ISPA58351.2023.10279445